fJscaler has started sampling a 100 GHz BiCMOS transimpedance amplifier (TIA) for intensity-modulation direct-detection (IM-DD) optical receiver designs aimed at AI and cloud data center interconnects. The company is pitching the part at a problem data center architects increasingly feel indirectly: as accelerator counts rise, optics power becomes a harder constraint on rack-level and facility-level power budgets.
The 100 GHz BiCMOS TIA is specified at 0.85 pJ/bit power efficiency, 18 pA/√Hz input-referred noise density, and 100 GHz bandwidth. fJscaler also says the device is optimized for operation with a back-illuminated InP photodiode. A next-generation version is in development targeting silicon photonics receivers.
fJscaler links the combination of sub-1 pJ/bit receiver energy and 100 GHz bandwidth to enabling 200G-per-lane and higher serial links, with what it describes as sensitivity headroom for low-power, unretimed, and linear-drive architectures. For data center operators, the practical takeaway is simple: shaving fractions of a picojoule per bit at the receiver can matter, but only if the full module (including DSP choices, photonics, and thermal design) lands where system designers need it.
“The economics of AI infrastructure are now written in picojoules per bit,” said Jan Filip, CEO of fJscaler. “With this TIA sampling to strategic customers today, we are giving module and system makers a receiver front-end that supports the 1.6T and 3.2T generations of optical interconnects without compromising on sensitivity or power.”
Alongside the TIA, fJscaler introduced a new family of Photonic IC (PIC) products targeting 1.6T coherent-lite and IM-DD applications. The company says the PIC lineup is intended to complement its electronic IC portfolio to support co-designed optical engines for data center interconnects, spanning intra-data center IM-DD links and longer-reach campus and inter-data center use cases using coherent-lite.
The 100 GHz BiCMOS TIA is sampling now to strategic customers. More information is available at fJscaler.
Source: fJscaler
